Re: [PATCH 1/2] hwmon: ina3221: Add power sysfs nodes

[Guenter Roeck]

Hi Nicolin,

On Wed, Sep 26, 2018 at 11:20:06AM -0700, Nicolin Chen wrote:
> On Wed, Sep 26, 2018 at 05:34:53AM -0700, Guenter Roeck wrote:
> > Hi Nicolin,
> > 
> > On 09/25/2018 11:42 PM, Nicolin Chen wrote:
> > > The hwmon sysfs ABI supports powerX_input and powerX_crit. This
> > > can ease user space programs who care more about power in total
> > > than voltage or current individually.
> > > 
> > > So this patch adds these two sysfs nodes for INA3221 driver.
> 
> > Ah, sorry, we can't do that. The sysfs nodes are for chips providing power
> > registers, not for kernel drivers to provide calculations based on voltage
> > and current measurements.
> 
> Hmm..I saw ina2xx.c and ltc4215.c are doing similar calculations...
> 

ina2xx.c doesn't; the chips supported by the driver do have a register
reporting the power (0x03). ltc4215.c was not reviewed by a hwmon
maintainer. I think I mentioned before that you can find anything you want
in the Linux kernel. That doesn't make it right.

> > Basic guideline is that we report what is there, not some calculation based
> > on it.
> 
> I could feel the back thoughts behind the guideline, but this does
> give user space programs some trouble -- I have a few programs that
> were used to read ina2xx driver which provides power nodes, but now
> those programs will have to implement another function to read the
> voltage and current separately to do further calculations.
> 
> Do you know any better solution for this situation?
> 

Userspace simply must not assume that power attributes exist and calculate
it from voltage and current attributes if needed.

> > This is even more true for power limits: We can not assume that the power limit
> > is (max voltage * max current). or (current voltage * max_current), or anything
> > else. We simply don't have the knowledge to make that assumption.
> 
> I agree that power limit is a bit tricky here as the voltage could
> change depending on the user space, Yes, I assumed that users who
> set_power() should be aware of it (whether fixed or dynamical) so
> as to decide to configure power limit or just current limit.
> 

You just can not make any such assumptions, sorry. Limit attributes
absolutely must be reflected in hardware. 

Thanks,
Guenter